Project Overview
Jarvis is a self-hosted AI assistant web application running on a home NAS. It provides a browser-based conversational interface powered by locally hosted large language models via Ollama, with voice input and an animated orb UI. The application is publicly accessible behind Laravel authentication.
Key Features
- Wake-word detection ("Hey Jarvis") and push-to-talk voice input in the browser
- Animated orb UI with distinct idle, listening, thinking, and speaking states
- Integration with local Ollama LLMs for private, on-device AI inference
- Text-to-speech response playback with XTTS voice synthesis
- Six informational widget tiles for contextual at-a-glance data
- Laravel authentication protecting the assistant interface
Technical Highlights
The frontend uses the Web Speech API for wake-word and push-to-talk voice capture. Responses are streamed from a local Ollama instance running on a separate GPU machine, keeping all AI inference private and off third-party servers. TTS playback uses XTTS for a natural voice with an edge-tts fallback.